Pith. sign in

REVIEW 4 major objections 7 minor 20 references

Large Models Enabled Ubiquitous Wireless Sensing

T0 review · 4 major / 7 minor · reviewed 2026-08-12 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read This paper claims that feeding a model ray-traced building geometry, not just raw channel data, makes spatial CSI prediction accurate enough for proactive beamforming.

desk verdict A plausible map-aware CSI prediction idea is undermined by an evaluation that never separates interpolation from generalization, and the paper's own numbers contradict the 'large models' framing. read the letter →

arxiv 2411.18277 v1 pith:I627Q6HX submitted 2024-11-27 cs.LG

Shun Hu This is my paper · ORCID
classification cs.LG
keywords CSIMIMO-OFDMspatialpredictionlargemodelswirelesscommunicationmachinelearningdata-drivenapproaches5G
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper tries to establish that spatial channel state information (CSI) in MIMO-OFDM systems, the matrix describing how signals travel from base station to user, can be predicted at unmeasured locations by combining a machine-learning predictor with physical knowledge of the environment. The route is to simulate realistic urban channels with ray tracing over a digital city map, then train models to map a 3D position to the channel matrix. The paper reports that a variational autoencoder reaches a validation NMSE of 0.0445, well below the no-prediction baseline around 0.5, while GPT-2, Transformer, and MLP baselines stay near 0.49. The author concludes that incorporating physical path characteristics reduces the uncertainty of the learned mapping and that the methodology should transfer to other urban maps. If this works in the field, base stations could anticipate channel conditions at locations users have not yet occupied, enabling proactive beamforming and resource allocation.

What carries the argument

The load-bearing mechanism is the map-aware feature-fusion preprocessing pipeline: raw geographic layout and TX/RX positions are turned into aligned tensors by embedding positions into the map, extracting hierarchical features with a CNN, concatenating multiscale intermediate features, encoding RX positions with a positional encoding module, and decomposing CSI into amplitude and phase. The predictor then learns the mapping from these aligned features to the channel matrix, trained with a smooth L1 loss. The paper argues that this physical grounding shrinks the function the model must learn, which is why accuracy improves: the environment determines the multipath geometry, so the model does not have to infer it from CSI statistics alone.

What would settle it

A direct test: train the VAE-based predictor on the ray-traced campus dataset and evaluate it on real measured CSI taken at the same grid of locations, or on a second city's ray-traced map. If validation NMSE jumps toward the no-prediction level around 0.5, the transfer claim fails. A second, already available check is the paper's Table IV: GPT-2 and Transformer validation NMSE around 0.49 means the large-model claim is false unless those models are evaluated on a task that plays to sequence modeling.

Watch

Extended reading notes

Core claim

On its own terms, the paper's central discovery is that a spatial CSI prediction framework can be built by fusing geographic features with channel data: transmitter and receiver positions are embedded into a map, the map is processed by a CNN to extract multiscale environmental features, receiver positions are positionally encoded, and CSI is decomposed into amplitude and phase before being fed together with the encoded positions into the predictor. Under this pipeline, the variational autoencoder predicts CSI across a ray-traced urban area with validation NMSE as low as 0.0445, an order of magnitude better than the large-sequence models (GPT-2 and Transformer) that plateau around 0.49 and the MLP baseline near 0.49. The paper also reports that the diffusion model reaches 0.125 with early stopping. The stated conclusion is that large models, when supplied with physical path characteristics, provide an effective solution for spatial CSI prediction, and that the likelihood of generalization to different and more complex urban maps is high.

Load-bearing premise

The load-bearing premise is that ray-traced simulated channels from a single digital urban map are a faithful stand-in for real physical channels, so that a model trained on those simulated maps will generalize to other maps and to real deployments.

Editorial extensions

If this is right

  • If spatial CSI prediction at unmeasured points is accurate, a network can form beams toward a user's future location before the user sends a pilot, reducing feedback overhead in 5G and 6G systems.
  • A model trained once on one ray-traced urban map could, under the transfer claim, be applied to new cities without per-site channel measurement campaigns.
  • The large gap between VAE and GPT-2 or Transformer indicates that for this static spatial task, generative latent-variable models are better suited than sequence-oriented large language models.
  • The success of the fused-feature pipeline suggests that the main gain comes from environmental feature engineering, so further gains may come from richer geometry such as materials, vegetation, and 3D building shapes.
  • Combining the spatial predictor with the time-series predictor from the LLM4CP reproduction, which reaches NMSE near 0.02 at low speeds, could give joint spatial-temporal CSI prediction.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The paper's own Table IV makes an implicit point the author does not draw out: the two models the conclusion credits as 'large models' (GPT-2 and Transformer) perform barely above a constant predictor, so the demonstrated success is specifically the VAE and diffusion model, not large language models as a class.
  • Because the ray-tracing simulation covers a single campus map with no real-world measurements, the strong generalization claim is untested; a natural next experiment is to train on one city map and evaluate on a second map or on measured channels, comparing NMSE to the no-prediction level.
  • The static ray-tracing setup removes the temporal dimension entirely, while the earlier QuaDRiGa section shows multi-time-slice inputs improved NMSE from 0.17 to 0.035; feeding time-sliced CSI into the map-aware spatial model may combine the two gains.
  • The wall-distance-and-orientation feature representation was abandoned for the full-map CNN; an ablation comparing those two feature sets would isolate how much geometric detail actually drives the VAE's accuracy.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 7 minor

Summary. The paper proposes a machine-learning framework for spatial CSI prediction in MIMO-OFDM systems, combining ray-tracing data from a realistic map (CUHK(SZ) via WinProp) with a feature-fusion architecture that feeds environmental, positional, and CSI information into several models (GPT-2, VAE, Transformer, Diffusion, MLP). It also reproduces a temporal CSI prediction experiment (LLM4CP) using QuaDRiGa. The central claim is that large models, augmented with physical environment features, can predict spatial CSI at unmeasured positions accurately enough to support proactive beamforming and resource allocation.

Significance. If the central claim were established, the work would be a useful step toward environment-aware CSI prediction for 5G/6G. The paper has some strengths: it uses a deterministic ray-tracing simulator (WinProp) on a real OpenStreetMap layout, reports a standard NMSE metric, and compares several architectures. The reproduction of the LLM4CP temporal prediction experiment is a reasonable benchmarking exercise. However, the primary scientific claim is not supported by the paper's own data. The VAE (which is not a 'large model' in the language-model sense) is the only model that achieves low validation NMSE, while GPT-2 and Transformer are essentially at the level of the MLP baseline. Moreover, the evaluation protocol does not establish spatial generalization because no spatially disjoint train/validation split is described, and the proposed feature fusion is never tested with an ablation. These issues are load-bearing for the paper's stated contribution.

major comments (4)
  1. [§VII.D, Table IV] The central 'large models' claim is contradicted by the paper's own results. GPT-2 and Transformer validation NMSE remain near 0.49 across all epochs, whereas the VAE reaches 0.0445 and the Diffusion model reaches 0.125. The VAE is not a language model, and the GPT-2/Transformer results are no better than the MLP baseline. The conclusion in §VIII that 'the use of large models like GPT-2 and VAE provide an effective solution' is therefore not supported by Table IV. The authors need to either reframe the claim to 'variational autoencoders' or provide evidence that a language model can actually perform spatial CSI prediction.
  2. [§VII.A, VII.C] The paper never states how the training and validation sets were split. Section VII.A places UEs on a uniform grid with 0.1 m spacing. If the split is random over individual grid points, then every validation point has training points only 0.1 m away, making the regression problem essentially interpolation at sub-6 GHz coherence distances. Under that protocol, the VAE's low validation NMSE (Table IV) does not demonstrate prediction at unmeasured positions; it may reflect memorization of near-duplicate samples. The authors must describe the split explicitly and, ideally, use a spatially disjoint holdout (e.g., a contiguous geographic zone or a coarser grid) to demonstrate genuine spatial generalization.
  3. [§VII.C, VIII] The claim in §VIII that incorporating physical characteristics of signal paths 'reduces the uncertainty of the function to be learned' is not tested. No ablation compares the full pipeline with a version that omits the environmental features, or compares the 'top 5 nearest walls' representation with the CNN-based map feature extraction. Without such an ablation, the reported gains cannot be attributed to the proposed feature fusion rather than to the choice of model architecture or data representation.
  4. [§VII.A, VIII] The method is evaluated on a single simulated environment (CUHK(SZ) via WinProp). Section VIII asserts the method 'is believed to be adaptive to totally different and complex urban maps,' but no second map, no real-world channel measurement, and no out-of-distribution test is provided. The paper should either add such a test or clearly state that cross-environment generalization is an open question. The simulation-to-reality transfer is assumed, not demonstrated.
minor comments (7)
  1. [§II] Several references are cited with bare arXiv links and without venue or year; e.g., [5], [6], [8], [9], [13], [15]. The citation style should be made consistent, and missing bibliographic details should be added.
  2. [Eq. (1)] The channel matrix notation in Eq. (1) is confusing: the entries h11, h12, ..., hn1, hnm suggest a rectangular arrangement that does not match the standard MIMO channel matrix notation h_{ij} with i as the receive index and j as the transmit index. Please correct the indices.
  3. [§V.C] The sentence 'Both frequency domain and delay domain information are feeded into the model' contains a typo ('feeded') and is vague about the actual input representation. Please specify the tensor shapes and the preprocessing steps.
  4. [§VI.B] The reproduced QuaDRiGa spatial prediction results report NMSE values (0.17 and 0.035) without an accompanying table or figure. Please add the corresponding learning curves or at least a table with standard deviations.
  5. [§VII.C] The initial approach using 'top 5 nearest walls' is described but no results are shown for this variant. Either present a comparison or remove the description to avoid an untested claim.
  6. [§VII.D] The figure captions contain typos: 'The learning cutve of Diffusion model' should be 'curve', and the text mentions Fig. 9, Fig. 10, and Fig. 11 but the figures are not included in the manuscript text as submitted.
  7. [Abstract] The abstract says 'language models' but the best-performing model in the spatial prediction task is a VAE, which is not a language model. Please align the abstract terminology with the actual models used.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: the CSI-prediction pipeline is evaluated on simulator-generated held-out positions, and no load-bearing self-citation or fitted-input-as-prediction step is present.

full rationale

The paper's central claim is empirical rather than derived: WinProp ray tracing generates CSI ground truth for positions in a CUHK(SZ) map, and machine-learning models are trained on position/environment features and evaluated with NMSE on validation positions (Sec. VII, Table IV). The prediction target is produced by an external simulator, not by the model or by the paper's own fitted parameters, so the evaluation does not reduce by construction to the model's inputs. No self-citations are load-bearing: references [1]-[19] contain no work by the sole author, and the reproduced LLM4CP baseline [14] is an independent prior framework. The paper does flag a real limitation in Sec. VI.C that "current implementation uses a statistical channel" with random scatter points, and the Sec. VII.A statement that "the distance between UEs in the grid is 0.1m" raises a legitimate concern about whether the train/validation split is spatially disjoint; similarly, the claim in Sec. VIII that incorporating physical path characteristics "reliably raises the accuracy" is not isolated by an ablation. However, these are correctness and external-validity risks, not circularity: there is no equation in which the predicted quantity is identical to the input by definition, no fitted parameter renamed as a prediction, and no uniqueness or ansatz imported through self-citation. The observed mismatch between the conclusion's "large models like GPT-2 and VAE" framing and Table IV's much lower VAE NMSE (0.0445) versus GPT-2 (0.4871) is an interpretive or reporting issue, not a circular reduction. Accordingly, no specific circular step can be exhibited, and the appropriate circularity score is 0.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The paper introduces no new physical entities. It combines existing simulators, QuaDRiGa and WinProp, with existing model architectures. The load-bearing items are simulation-fidelity assumptions and the untested feature-fusion hypothesis.

free parameters (4)
  • Top-5 nearest walls per UE = 5
    Hand-chosen count for the first geometric feature representation in Section VII.C. No ablation on this number is reported.
  • UE grid spacing = 0.1 m
    Chosen simulation density for WinProp ray tracing in Section VII.A. It affects the spatial correlation of training samples.
  • Diffusion model early-stop epoch = 30
    Early stopping after epoch 30 because NMSE no longer improves in Section VII.D. The checkpoint selection is post hoc.
  • ML model hyperparameters = not reported
    Architecture dimensions, learning rates, batch sizes, and optimizer settings for GPT-2, VAE, Transformer, diffusion, and MLP are not given, so the reported NMSE values cannot be reproduced or audited.
assumptions (5)
  • domain assumption Uplink and downlink CSI share statistical correlation due to channel reciprocity.
    Used in Section V.B to justify LLM4CP time-series prediction. Not validated with data in this paper.
  • domain assumption The 3GPP 38.901 Urban Macro model implemented in QuaDRiGa represents real-world mobile channels.
    Section V.C.1 uses this standardized scenario to generate training data for all models.
  • domain assumption WinProp ray tracing with OSM-derived geometry produces realistic CSI for spatial prediction and supports generalization to other maps.
    Section VII.A and Section VIII claim realism and generalizability, but no second map or real-world validation is provided.
  • ad hoc to paper Fusing environmental features with CSI improves prediction accuracy.
    Core hypothesis of the proposed framework in Section VII.C, but no ablation compares models with and without environment features.
  • ad hoc to paper Feeding positions and the raw map separately causes a degenerate effect in feature learning.
    Section VII.C states this as a design motivation for concatenating aligned features; no experiment demonstrates the degeneracy.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Large Models Enabled Ubiquitous Wireless Sensing." pith.science (2026). https://pith.science/paper/I627Q6HX

@misc{pith2026241118277,
  author       = {Pith},
  title        = {Pith review of: Large Models Enabled Ubiquitous Wireless Sensing},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/I627Q6HX}},
  note         = {Machine review of arXiv:2411.18277}
}
read the original abstract

In the era of 5G communication, the knowledge of channel state information (CSI) is crucial for enhancing network performance. This paper explores the utilization of language models for spatial CSI prediction within MIMO-OFDM systems. We begin by outlining the significance of accurate CSI in enabling advanced functionalities such as adaptive modulation. We review existing methodologies for CSI estimation, emphasizing the shift from traditional to data-driven approaches. Then a novel framework for spatial CSI prediction using realistic environment information is proposed, and experimental results demonstrate the effectiveness. This research paves way for innovative strategies in managing wireless networks.

Figures

Figures reproduced from arXiv: 2411.18277 by the authors.

Figure 1
Figure 1. Modern classification of wireless channels [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Numerical Methods of Winprop, with more approximation introduced [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 4
Figure 4. The LLM4CP and other baselines’ NMSE performance in relation to [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figures from the paper (6 more)
Figure 5
Figure 5. Figure 5: The LLM4CP and other baselines’ NMSE performance in relation to [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 6
Figure 6. Figure 6: Procedure from physical environment buil-up, channel generation, to [PITH_FULL_IMAGE:figures/full_fig_p005_6.png]
Figure 8
Figure 8. Figure 8: Novel architecture of machine learning targeted at informing CSI ac [PITH_FULL_IMAGE:figures/full_fig_p006_8.png]
Figure 7
Figure 7. Figure 7: (a) The selected region of CUHK(SZ) as the place for ray-tracing [PITH_FULL_IMAGE:figures/full_fig_p006_7.png]
Figure 11
Figure 11. Figure 11: Learning curves of GPT-2, Transformer and MLP [PITH_FULL_IMAGE:figures/full_fig_p007_11.png]
Figure 10
Figure 10. Figure 10: The learning cutve of Diffusion model VIII. CONCLUSION In this paper, the study starts with probablistic modeling like QuaDRiGa, and switches to a more robust, physics-driven approach. By integrating raytracing-based channel generation technique, we successfully creat…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

20 extracted references · 16 canonical work pages

  1. [1]

    Spatial Channel State Information Prediction with Generative AI: Towards Holographic Communication and Digital Radio Twin

    L. Zhang, H. Sun, Y . Zeng, and R. Q. Hu, ”Spatial Channel State Information Prediction with Generative AI: Towards Holographic Com- munication and Digital Radio Twin,” arXiv:2401.08023, 2024

    work page Pith review arXiv 2024

  2. [2]

    Al-Asadi, I

    A. Al-Asadi, I. R. K. Al-Saedi, S. K. Alwane, H. Li, and L. Alzubaidi, ”Enhanced MIMO CSI Estimation Using ACCPM with Limited Feed- back,” in Sensors, vol. 23, no. 18, p. 7965, Sep. 2023. [Online]. https://doi.org/10.3390/s23187965

    work page doi:10.3390/s23187965 2023

  3. [3]

    Gheryani, ”A new approach to the design of adaptive MIMO wireless communication systems,” 2009

    M. Gheryani, ”A new approach to the design of adaptive MIMO wireless communication systems,” 2009. https://core.ac.uk/download/211515877.pdf

    work page arXiv 2009

  4. [4]

    A. M. Abd El-Moaty and A. Zerguine, ”Sparse Channel Estima- tion with Gradient-Based Algorithms: A comparative Study,” 2018. https://arxiv.org/pdf/1812.04196

    work page Pith review arXiv 2018

  5. [5]

    Z. Liu, L. Zhang, and Z. Ding, ”Overcoming the Channel Esti- mation Barrier in Massive MIMO Communication Systems,” 2019. https://arxiv.org/pdf/1912.10573

    work page Pith review arXiv 2019

  6. [6]

    H. Kim, S. Kim, H. Lee, C. Jang et al., ”Massive MIMO Chan- nel Prediction: Kalman Filtering vs. Machine Learning,” 2020. https://arxiv.org/pdf/2009.09967

    work page Pith review arXiv 2020

  7. [7]

    Pecorella, R

    T. Pecorella, R. Fantacci, and B. Picano, ”Improving CSI Prediction Accuracy with Deep Echo State Networks in 5G Networks,” 2020. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7697607/

    work page 2020

  8. [8]

    Zhang, A

    Y . Zhang, A. Alkhateeb, P. Madadi, J. Jeon et al., ”Predicting Future CSI Feedback For Highly-Mobile Massive MIMO Systems,” 2022. https://arxiv.org/pdf/2202.02492

    arXiv 2022

Show all 20 references
  1. [9]

    B. Ko, H. Kim, and J. Choi, ”Massive MIMO Channel Prediction Using Machine Learning: Power of Domain Transformation,” 2022. https://arxiv.org/pdf/2208.04545

    2022 arXiv

  2. [10]

    Park and O

    S. Park and O. Simeone, ”Speeding up Training of Linear Predictors for Multi-Antenna Frequency-Selective Channels via Meta-Learning,” 2022. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9600732/

    2022

  3. [11]

    H. Kim, J. Choi, and D. J. Love, ”Massive MIMO Channel Prediction Via Meta-Learning and Deep Denoising: Is a Small Dataset Enough?,”

  4. [12]

    T. Zhou, H. Zhang, B. Ai, C. Xue and L. Liu, ”Deep- Learning-Based Spatial–Temporal Channel Prediction for Smart High-Speed Railway Communication Networks,” in IEEE Transactions on Wireless Communications, vol. 21, no. 7, pp. 5333-5345, July 2022, doi: 10.1109/TWC.2021.3139384...

    2022

  5. [13]

    Y . Du, S. Chang Liew, K. Chen, and Y . Shao, ”The Power of Large Language Models for Wireless Communication System Development: A Case Study on FPGA Platforms,” 2023. https://arxiv.org/pdf/2307.07319

    2023 arXiv

  6. [14]

    L. Liu, X. Liu, S. Gao, X. Cheng, and L. Yang, ”LLM4CP: Adapting Large Language Models for Channel Prediction,” arXiv:2406.14440, 20 Jun. 2024

    2024 arXiv

  7. [15]

    H. Zhou, C. Hu, Y . Yuan, Y . Cui et al., ”Large Language Model (LLM) for Telecommunications: A Comprehensive Survey on Principles, Key Techniques, and Opportunities,” 2024. https://arxiv.org/pdf/2405.10825

    2024 arXiv

  8. [16]

    Q. F. Channel Simulators for mmWave and 5G Applications. 2017. https://api.semanticscholar.org/CorpusID:56454171

    2017

  9. [17]

    WINNER II Channel Models,

    P. Ky ¨osti, J. Meinil ¨a, L. Hentil ¨a, X. Zhao, T. J ¨ams¨a, C. Schneider, M. Narand ˇzi´c, M. Milojevi ´c, A. Hong, J. Ylitalo, V .-M. Holappa, M. Alatossava, R. Bultitude, Y . de Jong, and T. Rautiainen, “WINNER II Channel Models,” 2008, https://www.cept.org/files/8339/win...

    2008

  10. [18]

    Fraunhofer Heinrich Hertz Institute, Wireless Communications and Networks, Quasi Deterministic Radio Channel Generator User Manual and Documentation, Document Revision: v2.8.1, Einsteinufer 37, 10587 Berlin, Germany, December 13, 2023

    2023

  11. [19]

    MIMO Simulation in 5G Networks: Py5cheSim and DeepMIMO In- tegration

    S ´anchez, D., Trujillo, M., Varela, P., Rattaro, C., Ingl´es, L., & Belzarena, P. MIMO Simulation in 5G Networks: Py5cheSim and DeepMIMO In- tegration. In 2023 XLIX Latin American Computer Conference (CLEI) (pp. 1-7). 2023

    2023

  12. [2022]

    https://arxiv.org/pdf/2210.08770

    arXiv

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.